The lammps can not produce the right structures using vasp MLFF as the fast model

[suojiang_zhang1]

Dear,
I still want to run the third part MD simulation by vasp MLFF to accelerate the calculation. I am glad to see you provide lammps interface to establish the communication. I try to use it, but I am not successful ideally to train the liquid system (maybe the solid is better)
Using the same MLFF after refit, I compared the results of vasp fast model ( ML_MODE=run) and lammps, I found that the fast model is good and the trajectory structures keep chemical bonding (figure 1) ,but the lammps structures show the bond broken and non-chemical meaning bonding at short time (figure 2), moreover, the running speed is very slower than the fast model.